From 80d1b5f5436082f0ae11d3739f53f7e7b872671e Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Constantin=20F=C3=BCrst?= <c@fuersten.info>
Date: Mon, 27 Nov 2023 11:36:36 +0100
Subject: [PATCH] remove fractional average calculation in favour of
 std::accumulate and remove option to set thread affinity to core while
 keeping support for node affinity assignment

---
 benchmarks/benchmark.hpp         | 30 ++++++++----------------------
 benchmarks/task-data.hpp         |  7 ++-----
 benchmarks/task-description.json |  3 +--
 3 files changed, 11 insertions(+), 29 deletions(-)

diff --git a/benchmarks/benchmark.hpp b/benchmarks/benchmark.hpp
index 7b736ff..a92ba79 100644
--- a/benchmarks/benchmark.hpp
+++ b/benchmarks/benchmark.hpp
@@ -3,6 +3,7 @@
 #include <iostream>
 #include <vector>
 #include <chrono>
+#include <numeric>
 
 #include <pthread.h>
 #include <semaphore.h>
@@ -13,39 +14,24 @@
 #include "statuscode-tostring.hpp"
 #include "task-data.hpp"
 
-double avg(const std::vector<double>& v) {
-    int n = 0;
-    double mean = 0.0;
-
-    for (const auto x : v) {
-        const double delta = static_cast<double>(x) - mean;
-        mean += delta / ++n;
-    }
-
-    return mean;
+double avg(const std::vector<uint64_t>& v) {
+    return static_cast<long double>(std::accumulate(v.begin(), v.end(), 0)) / static_cast<long double>(v.size());
 }
 
 #define LOG_CODE_INFO "Location: " << __FILE__ << "@" << __LINE__ << "::" << __FUNCTION__ << std::endl
-#define LOG_ERR { pthread_t t = pthread_self(); std::cerr << "--- BEGIN ERROR MSG ---" << std::endl << "Physical: [Node " << args->numa_node << " | Core " << args->core << " | Thread " << t << "]" << std::endl; } std::cerr << LOG_CODE_INFO
-#define CHECK_STATUS(status,msg) { if (status != dml::status_code::ok) { LOG_ERR << "Status Code: " << StatusCodeToString(status) << std::endl << #msg << std::endl; args->status = status; return nullptr; }}
+#define LOG_ERR { pthread_t t = pthread_self(); std::cerr << "--- BEGIN ERROR MSG ---" << std::endl << "Physical: [Node " << args->numa_node << " | Thread " << t << "]" << std::endl; } std::cerr << LOG_CODE_INFO
+#define CHECK_STATUS(status,msg) { if (status != dml::status_code::ok) { LOG_ERR << "Status Code: " << StatusCodeToString(status) << std::endl << ##msg << std::endl; args->status = status; return nullptr; }}
 
 template <typename path>
 void* thread_function(void* argp) {
     TaskData* args = reinterpret_cast<TaskData*>(argp);
 
-    std::vector<double> submission_durations;
-    std::vector<double> completion_durations;
-    std::vector<double> combined_durations;
+    std::vector<uint64_t> submission_durations;
+    std::vector<uint64_t> completion_durations;
+    std::vector<uint64_t> combined_durations;
 
     // set numa node and core affinity of the current thread
     numa_run_on_node(args->numa_node);
-    cpu_set_t cpuset;
-    CPU_ZERO(&cpuset);
-    CPU_SET(args->core, &cpuset);
-    if (pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset) != 0) {
-        LOG_ERR << "Error setting affinity for thread" << std::endl;
-        return nullptr;
-    }
     
     // allocate memory for the move operation on the requested numa nodes
     void* src = numa_alloc_onnode(args->size, args->nnode_src);
diff --git a/benchmarks/task-data.hpp b/benchmarks/task-data.hpp
index 3a0b74d..273a094 100644
--- a/benchmarks/task-data.hpp
+++ b/benchmarks/task-data.hpp
@@ -9,7 +9,6 @@
 struct TaskData {
     // thread placement / engine selection
     uint8_t numa_node;
-    uint8_t core;
     // region size and source+destination for move
     size_t size;
     uint8_t nnode_src;
@@ -38,8 +37,7 @@ inline void to_json(nlohmann::json& j, const TaskData& a) {
     j["task"]["batching"]["enabled"] = a.batch_submit;
     j["task"]["batching"]["batch_size"] = a.batch_size;
     j["task"]["batching"]["barrier_after_n_operations"] = a.barrier_after_n_operations;
-    j["affinity"]["cpu_core"] = a.core;
-    j["affinity"]["numa"] = a.numa_node;
+    j["affinity"]["node"] = a.numa_node;
     j["affinity"]["nnode_src"] = a.nnode_src;
     j["affinity"]["nnode_dst"] = a.nnode_dst;
     j["time"]["unit"] = "microseconds";
@@ -56,8 +54,7 @@ inline void from_json(const nlohmann::json& j, TaskData& a) {
     j["task"]["batching"]["enabled"].get_to(a.batch_submit);
     j["task"]["batching"]["batch_size"].get_to(a.batch_size);
     j["task"]["batching"]["barrier_after_n_operations"].get_to(a.barrier_after_n_operations);
-    j["affinity"]["numa"].get_to(a.numa_node);
-    j["affinity"]["core"].get_to(a.core);
+    j["affinity"]["node"].get_to(a.numa_node);
     j["affinity"]["nnode_src"].get_to(a.nnode_src);
     j["affinity"]["nnode_dst"].get_to(a.nnode_dst);
 }
diff --git a/benchmarks/task-description.json b/benchmarks/task-description.json
index 0fd0ca5..1172f8b 100644
--- a/benchmarks/task-description.json
+++ b/benchmarks/task-description.json
@@ -15,8 +15,7 @@
         }
       },
       "affinity": {
-        "numa": 0,
-        "core": 0,
+        "node": 0,
         "nnode_src": 0,
         "nnode_dst": 0
       }